Autoimmune diseases – such as type 1 diabetes (T1D), systemic lupus erythematosus (SLE), and multiple sclerosis (MS) – affect millions worldwide. Although these diseases manifest differently – attacking distinct tissues like the pancreas in T1D, connective tissues and organs in SLE, or the nervous system in MS – these disorders share a surprising common thread: strong genetic associations within the human leukocyte antigen (HLA) system. HLA class II proteins normally show fragments of proteins (peptides) to immune cells, helping the body distinguish friend from foe. But certain HLA variants appear to present “self-peptides” in harmful ways, triggering chronic immune attacks on the body’s own tissues. To uncover why some HLA alleles increase risk while others offer protection, we will use a cutting-edge yeast display platform. This approach lets us systematically screen millions of human protein fragments, revealing which ones latch onto “risk” versus “protective” HLA molecules. Next, we will apply advanced computer models—called protein language models—to understand the structural “grammar” of these interactions. By doing so, we can predict which peptides are most likely to incite damaging immune responses or promote tolerance. Finally, we will delve into single-cell RNA sequencing data to pinpoint the tissues and cell types that generate these critical peptides, offering insight into why the immune system might become overactive in T1D, SLE, or MS. Ultimately, our work aims to move beyond generic immune suppression by spotlighting the precise peptides that drive autoimmunity – or block it. Such knowledge will pave the way for highly targeted therapies that interrupt pathogenic pathways without compromising healthy immune function. In clarifying the role of selfpeptides in disease, we will chart a path toward more personalized and effective treatments that benefit patients across a range of autoimmune disorders.

Autoimmune diseases, including T1D, SLE, and MS, consistently show strong associations with specific human leukocyte antigen (HLA) class II alleles. While these risk alleles (e.g., DQB103:02 in T1D, DRB103:01 in SLE, DRB115:01 in MS) are known to present self-antigens in ways that promote pathological immune responses, protective alleles (e.g., DQB106:02 in T1D) appear to foster tolerance through largely unknown mechanisms. Deciphering precisely which self-peptides these alleles present – and how this presentation differs between risk and protective variants – remains a pivotal unmet need in autoimmune research. Our approach combines high-throughput screening, computational modeling, and single-cell transcriptomics to address this gap. First, we will use a yeast display platform to scan a 49-mer peptide library comprehensively spanning the human proteome. This library ensures overlapping coverage of diverse self human protein fragments, allowing us to capture potential autoantigenic epitopes. Engineered yeast cells expressing linked HLA class II variants will undergo fluorescence-activated cell sorting (FACS) to isolate high-affinity binders, followed by next-generation sequencing (NGS) of recovered peptide barcodes. This screening will generate a detailed map of self-peptides that preferentially bind risk versus protective HLA alleles. Next, we will apply protein language models tailored for peptide-HLA interactions. These transformer-based architectures will integrate sequence and structural features to classify peptides as likely “risk” or “protective.” Structural modeling will further refine these predictions by delineating how peptide orientation and binding-groove fit may promote immunogenicity or tolerance. Finally, we will analyze single-cell RNA sequencing (scRNA-seq) datasets from disease-relevant tissues such as pancreatic islets for T1D or central nervous system samples for MS – to determine which cell populations express the proteins yielding critical peptides. By correlating expression data with HLA binding profiles, we will pinpoint tissue-specific mechanisms that underlie pathogenic versus protective immune responses. Collectively, these studies will yield (1) a high-confidence atlas of self-peptides associated with HLAmediated risk and protection; (2) predictive computational models to identify pathogenic versus tolerogenic motifs; and (3) context-specific insights into where these peptides originate in vivo. Ultimately, this knowledge will guide the design of precise interventions -such as peptide-specific blockers or engineered tolerogenic pathways – offering a targeted alternative to broad immunosuppression.